Information Security Journal: A Global Perspective

ISSN: 1939-3555 (Print) 1939-3547 (Online) Journal homepage: www.tandfonline.com/journals/uiss20

Chaos based Medical Image Encryption

Techniques: A Comprehensive Review and
Analysis

Shubha Pankaj & Mohit Dua

To cite this article: Shubha Pankaj & Mohit Dua (2024) Chaos based Medical Image Encryption
Techniques: A Comprehensive Review and Analysis, Information Security Journal: A Global
Perspective, 33:3, 332-358, DOI: 10.1080/19393555.2024.2312975

To link to this article: https://doi.org/10.1080/19393555.2024.2312975

Published online: 13 Feb 2024.

Submit your article to this journal

Article views: 501

View related articles

View Crossmark data

Citing articles: 12 View citing articles

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=uiss20
INFORMATION SECURITY JOURNAL: A GLOBAL PERSPECTIVE
2024, VOL. 33, NO. 3, 332–358
https://doi.org/10.1080/19393555.2024.2312975

Chaos based Medical Image Encryption Techniques: A Comprehensive Review

and Analysis
Shubha Pankaj and Mohit Dua
Department of Computer Engineering, National Institute of Technology, Kurukshetra, India

ABSTRACT KEYWORDS
In the past two decades, we have seen the proliferation of Internet and its applications. The Chaos; cryptography; DNA;
information exchange happens via images, videos, texts, audio clips, etc. Medical experts use image encryption; security
images as the most common form of information exchange such as X-rays, MRI scans, CT scans, etc.
These scanned images contain confidential information about a patient. With images being the
most attractive mode for medical information exchange, it becomes the target of many security
attacks. Many researchers have put out their best efforts to protect the images from these attacks
by proposing several techniques. Despite all efforts, many cryptosystems still compromise with one
or more key attributes like efficiency or time required for them to be top-notch. Along with an
appreciation of work, careful inspection is also required that can influence the future work and
make them better. Hence, this survey aims to summarize and analyze the existing chaos-based
medical image encryption techniques introduced in the last five years. The paper starts with a brief
introduction about the basic chaos maps, security parameters, and then, progresses to analysis of
the encryption algorithms based on categories followed by comparative study and open security
issues.

1. Introduction
addition or deletion, or modification in the original
With the exponential growth of Internet technolo­ data is accepted. There are two points where
gies by each day, data security becomes a serious Attackers can compromise the integrity of the
issue. If the internet can provide someone with data during upload or download. Hence, one
several facilities by giving up basic information, it must choose a reliable channel for transmission.
can also prove to be a threat using that informa­ If the receiver is authorized and wants some data,
tion. Hence, at every step of information exchange, the data shall be made available. This property
we require the mechanism’s security and self- stands for Availability.
awareness. For instance, many companies or start- These three pillars of data security also power
ups offer medical facilities, including medicines the concept of image security. For example, the
supply and laboratories that send crucial data of image should be secured so that the authorized
the patient over insecure channels connected by the parties can only access it. We can encrypt the
internet. If one can exploit the weakness of the data with some algorithm that can only be
insecure channel, they can get their hand over decrypted by using the correct keys. These
those data. These data can be sold to many insur­ valid keys should only be present with the
ance policy companies to analyze the data and force authorized party. Therefore, confidentiality is
patients to buy their policies. The main three pillars ensured. Similarly, if an unauthorized party
are shown in Figure 1 that establishes the concept tries to modify the data after encryption, the
of data security are Confidentiality, Integrity, and decrypted image should not result in the origi­
Availability. Confidentiality means that data nal encrypted image. Also, the encryption
should only be available to respective authorized scheme should be strict so it will not produce
parties. Therefore, the content is shared only any unwanted change during encryption and
between sender, receiver, and no one else. decryption. Availability is ensured if the security
Integrity stands for data being tamperproof. No scheme is always ready to work on the image.

CONTACT Mohit Dua [email protected] Department of Computer Engineering, National Institute of Technology, Kurukshetra, Haryana, India
© 2024 Taylor & Francis Group, LLC
 INFORMATION SECURITY JOURNAL: A GLOBAL PERSPECTIVE 333

state. Confusion does not modify the value of pixels

but changes their original position. Diffusion pro­
duced a reversible change in the pixel values; one
can say that diffusion masks the original pixel
values. To support these above-mentioned opera­
tions, image compression can be performed at pre-
processing or post-processing stages. However, out
of all encryption schemes, our focus is on chaos-
based cryptosystems.
In the next Section 2, we have covered funda­
mentals of chaos maps, and in Section 3,
a discussion on various performance evaluation
parameters is given. Section 4 categorizes the ana­
Figure 1. Three pillars of data security (Samonas & Coss, 2014).
lysis on the basis of techniques used. Section 5,
presents the comparative study. Section 6 points
We can secure images through many methods, out the open security issues, and brief summary of
as shown in Figure 2. For example, we have our image encryption schemes. Section 7 concludes our
traditional encryption schemes like Data study.
Encryption Standard (DES), Triple Data
Encryption Standard (TripleDES), Message Digest
Algorithm 5 (MD5), International Data Encryption 2. Chaos-based cryptography
Algorithm (IDEA), etc. (Devi et al., 2015; El-Deen A mathematical model showing random/chaotic
et al., 2014; Gupta et al., 2009; Laiphrakpam & behavior is considered a chaotic system. A chaos-
Khumanthem, 2017; Yun-Peng et al., 2009). There based system can be discrete or continuous. There
are many problems with these techniques; one is has been a noticeable trend of using chaos-based
high pixel correlation. Secondly, some of these encryption schemes in recent years because of
methods have repetitive key generation patterns, favorable properties (Kocarev et al., 1998) like ran­
making them vulnerable. Apart from these techni­ domness, extensive chaotic range, sensitivity to
ques, we have our modern approach to image initial conditions, and control parameters.
security. We can hide the data through watermark­ Another possibility for favoring chaos-based
ing. We can also apply the basic confusion- schemes is that it can combine already existing
diffusion principle to mix up the image’s original methods with it. For instance, we have

Figure 2. Image security techniques classification (Ahmad & Hwang, 2016; Akkasaligar & Biradar, 2016, 2020; Wolfgang & Delp, 1998).
334 S. PANKAJ AND M. DUA

a combination of chaos and watermarking, chaos

and compression, chaos and neural networks, etc.
Chaos-based cryptosystems systems use chaotic
functions or maps for implementing encryption. ● Chaotic Standard map
We have briefly mentioned some of the commonly
used chaotic maps below; kindly look for other It is defined by the equation:
maps too.

● Logistic map (LM)

here, K is the control parameter and is greater than
It is a polynomial mapping of degree 2 and given by 0. For K equals 0, the map is linear and only quasi­
the equation: periodic and periodic orbits are possible. Along with
the increase in value of K, non-linearity of the map
also increases. The size of chaotic system is repre­
sented by N. xn and yn lie in the range of 0 to 2π.
where μ lies in range of 0 to 4 and 0 < X0 < 1. n is
number of iterations. Xn describes the ratio ● Henon map
between the existing and the maximum possible
population. If μ gets over the range of 4, then, the Sometimes referred as Henon-Pomeau attractor
logistic tends to make populations size negative. If map (Marotto, 1979) is a discrete domain-
μ is between 1 and 2, then Logistic map becomes dynamic system. It takes one point (xn ; yn ) along
independent from the initial population. To keep the plane and then maps them to another point.
the bifurcation diagram at it’s best, μ is kept equal Mathematically it is defined as:
to 4 for most chaotic systems using Logistic map.
One can refer to (Hua & Zhou, 2016) for getting to
know more about this map.
where a and b are the parameters on which the map
● Sine map relies.
As mentioned earlier, researchers favor chaos-
Eq (2) shows the sine map’s mathematical based cryptosystems because, along with these
formulation: maps and their hybrid versions, we can combine
different techniques. One such technique is permu­
tation-diffusion. It decreases the correlation
between adjacent pixels and also introduces
where X lies in range of 0 to 1 and μ is greater a slight change in pixel values. Figure 3 shows the
than 0. general architecture of how confusion and diffu­
sion work in sequence. Suppose the image is of size.
● Tent map Then, in the classical scheme, we iterate the confu­
sion stage n times; after that, we apply diffusion
The tent map (Yoshida et al., 1983) is given by Eq once. This whole process is repeated m times.
(3) where μ is a real positive number. Xn lies in Another method is hyper-chaos with methods like
a range of 0 to 1. Generally, μ equals to 2 because if cellular automata or DNA. Neural Network (NN)
the value of μis less than 1 then the system has many layers, which helps encrypt images better.
approaches 0 for the x-axis values. If μ equals 1, Motivated by all the work done by researchers,
then some of points on x-axis become fixed for the we try to look into many proposed schemes. Most
system. For, μ >1, only two points become fixed. of the encryption schemes use more than one chao­
1<μ <2, plots both non-periodic and periodic tic map along with the techniques mentioned
points. If value is greater than 2, then the points above. These chaotic maps are of two types: low-
become disconnected sets. dimensional (LD) (Dou & Li, 2020; Ponuma &
 INFORMATION SECURITY JOURNAL: A GLOBAL PERSPECTIVE 335

an effective chaos map. Figure 4 defines the bifurca­

tion diagrams of the Logistic, Sine, and Tent map.

● Lyapunov Exponent

LE (Frank, 1990) measures the sensitivity of chaos

map with respect to the changes in initial conditions.
The divergence degree and rate of separation of tra­
Figure 3. Classical confusion-diffusion scheme. jectories that start their movement from an immen­
sely close initial point are indicated by LE. A good
chaos system should have positive LE. The fast diver­
Amutha, 2018) and high-dimensional (HD) gence of trajectories is mapped by greater LE values.
(Hosseinzadeh et al., 2019; Shokouh Saljoughi & Fast divergence implies greater chaotic behavior. For
Mirvaziri, 2019) chaotic maps. LD chaotic maps a chaotic system f ðyi ), the LE is given by:
have a basic structure that is one dimensional,
having one variable. These maps are easy to imple­
ment. Because LD has a small chaotic range, it
becomes easy to predict. In contrast, HD maps
are complicated in terms of implementation but ● Entropy
have an extensive chaotic range.
A positive entropy indicates that the chaos map is
performing good. This shows that repetitive pat­
3. Performance and security evaluation terns are absent. The larger ApEn values indicate
that the system is highly complex.
A good encryption scheme comprises an efficient
chaos map and an effective encryption scheme.
When checked for the Bifurcation diagram,
3.2. Performance and security evaluation of
Lyapunov Exponent, and Entropy, the map should
encryption scheme
perform well. It should collectively be robust to all
known attacks like statistical, chosen-plaintext, ● Correlation Analysis
ciphertext, and exhaustive attacks. Figure 5 shows
all the commonly used performance and security In a plain image, adjacent pixels are strongly corre­
metrics that mark an algorithm’s good. Some of lated vertically (CCV), horizontally (CCH), and diag­
them are discussed next two subsections. onally (CCD). The minimum and maximum
correlation coefficients are 0 and 1, and for
a strongly encrypted image the correlation coefficient
3.1. Performance and security evaluation of chaos should be 0. The values are calculated as following:
map
● Bifurcation Diagram

This parameter visually analyses the behavior of

a chaos map (Bigge & Bohl, 1985). The objective is
to determine the periodicity, instability, and stabi­
lity. When the changes occur in the control para­
meters, the bifurcation diagram also changes. For
a chaos map to be good, uniformity in the bifurca­
tion diagram is a must. The white space represents
the chaotic range that the tested chaos map does not
cover. There should be no blank windows in case of
336 S. PANKAJ AND M. DUA

Figure 4. Bifurcation diagrams (a) logistic map (b) sine map (c) tent map.

Here, y and x are horizontal and vertical adjacent ● Entropy

pixels, N is the total number of pixels. covðx; yÞ is
covariance at pixel positions x and y. It is a measure of randomness and spread of pixels.
IE is given by equation:
● Histogram Analysis

It shows the pixel distribution. For a good encryption

algorithm, the histogram plotted on ciphered image where, symbol set is represented by s,2N 1 are the
should be uniform. Uniformity makes information total number of symbols si 2 s. Pðsi Þ represents the
unpredictable, and a histogram attack can be avoided. probability of random variable si . The Entropy of
 INFORMATION SECURITY JOURNAL: A GLOBAL PERSPECTIVE 337

Histogram
analysis

Sta!s!cal Correla!on
Analysis analysis

Entropy analysis

GVD analysis
Randomness
test
NIST analysis

Plain image
sensi!vity
Differen!al
analysis
Avalanche effect
Sensibility
analysis
Secret key space
analysis
Security key
Performance
analysis
and Security
Key sensi!vity
analysis
analysis
Contrast
adjustment
Gaussian noise
Noise analysis
interference
analysis Occlusion noise
analysis
Robustness Histogram
against noise equaliza!on

JPEG
Speed analysis
Compression

Encryp!on
Cropping a#ack
quality

Ciphertest-only
a#ack

Known-plaintext
Poten!al a#ack
a#ack

Chosen-plaintext
a#ack

Figure 5. Common performance and security metrics [17].

an encryption algorithm should be around eight for This indicates the average number of changes to
it to prevent statistical attacks. each gray level L. EQ is given by:

● Encryption Quality
338 S. PANKAJ AND M. DUA

where Eði; jÞ and I ði; jÞ are the gray value of the To test the robustness of a cryptosystem against the
pixels placed at position (i,j) in plain and ciphered noise produced during transmission or by the
images. HL ðI Þ and HL ðEÞ are defined as then num­ environment, noise with mean zero and variance
ber of occurrences for each gray level L in cipher and of 0.0001,0.0003 and 0.0005 is produced in the
plain image. For EQ to be good, it should be large. image. Also, salt and pepper noise is added with
density of 1%, 5%,10% and 25%.
● Sensitivity
● Histogram Equalization
It is evaluated by two parameters, number of pixel
change rate (NPCR) and unified average changing The parameters like Peak signal to noise ratio
intensity (UACI) (Wu et al., 2011). NPCR shows (PSNR) and mean square error (MSE) are used to
how small changes in the original image can affect check histogram equalization. MSE indicates
the final ciphered image. The difference between a collective squared error between original (P)
the average intensity of two images is given by and the ciphered image (C).
UACI. An encryption scheme producing higher
values of UACI is considered good. NPCR and
UACI are given by:

● Cipher text-only Attack

In this attack, one tries to decrypt the cipher text when

cipher text is already available. The attacker tries to
recover the decrypt the image or acquire the keys.

● Known plaintext Attack

The attacker has the associated plain text and

respective ciphered text and tries to get the keys.
● Avalanche Effect

This indicate the number of bits changed in cipher 4. Medical image cryptosystems classification
image when we change bits in original image. The
ideal value is 0.5. In the upcoming subsections, we have categorized
chaos-based image encryption algorithms based on
● Secret key space analysis different methods that they have used as shown in
Figure 6. Table 1 contains references of the papers
This is one of the most important feature for reviewed under each category. Based on Figure 7.
strength determination of an algorithm. The larger The number of articles using Miscellaneous meth­
the key space, the harder it is to break cryptosys­ ods, chaos-based, DNA-based, and Confusion-
tem. Doubling the key size does not mean that Diffusion is higher than that of others method.
number of required operations will also double,
rather it is squared. For a cryptosystem to be
good enough to withstand attacks, the key space 4.1. Only chaotic map based encryption algorithms
should be larger than 2128 . Although almost all the chaos cryptosystems use
the chaotic map as their base, we have placed those
● Gaussian Noise Analysis algorithms with a simple framework and no other
 INFORMATION SECURITY JOURNAL: A GLOBAL PERSPECTIVE 339

Figure 6. Categorization of chaos based medical encryption schemes.

underlying complex encryption method in this Sine map 2 are column-wise confused. Results and
subsection. both Sine maps and Rossler system are diffused
One such algorithm was proposed by Ibrahim row and column-wise to output the cipher image.
et al. (Ibrahim et al., 2020). They have used They have designed this multi-dimensional system
dynamic substitution S-boxes and 2-D chaotic to increase the keyspace size and sensitiveness
maps. AES encrypts the keys to provide additional toward it. This will protect the system from many
security, are fed into chaotic map and PRNG attacks.
(OISHI & Inoue, 1982). The PRNG generates the In (S. Kumar et al., 2019), Kumar et al. have
input for S-boxes. The algorithm starts with plain proposed a new scheme using Fraction discrete
image pixels getting XORed with the first instance cosine transform (FrDCT) (Cariolaro et al., 2002)
of S-Box. along with the chaotic map. They have selected
The result is again XORed with the chaotic FrDCT over classical Fourier Transform (FRFT)
sequence generated earlier. Again, this result is because FRFT produces both real and imaginary
XORed with the second instance of S-Box. This coefficients, which makes processing difficult. The
process continues until it reaches the chaotic tran­ algorithm begins by breaking the image into four
sient length (covering all pixels), finally giving the blocks that should not overlap. On these blocks,
ciphered image as output. The decryption is the FrDCT is applied separately. The four blocks are
inverse of the encryption algorithm. The keyspace joined together and passed on to the chaotic sys­
is relatively large because of the use of S-Boxes. tem. In the chaotic system consists of 2 matrices
In (Sangavi & Thangavel, 2020), Sangavi et al. and 1 array; Symbolized matrix S, Transformation
has proposed a multi-dimensional medical image matrix P, and Scrambling array H. Chebychev map
encryption scheme that employs Rossler dynamical is used to get the S matrix. The P matrix is obtained
system and Sine map. The proposed method with­ by mathematical double, mod, and round opera­
stands many statistical attacks and performs well. tions. The H array stores the index location and
The input image is divided into two vectors of blocks are arranged in a given manner. After pas­
equal sizes. The vectors are summed separately sing through the chaotic system, ciphered image is
and then multiplied, taken modulo to the image obtained.
twice the original image’s size. The quantization Ke et al. (Ke et al., 2019) proposed a four-step
process is carried out to remove the unwanted encryption process. The embedded secret message
signals on all vectors, Rossler system and Sine is replaced with MSB. Since during the data hiding
map system (1 & 2). Quantized Rossler system phase, the MSB value is lost, hence during the
and Sine map 1, along with vectors, are confused decoding phase, the correct value must be pre­
row-wise. Again, the quantized Rossler system and dicted. The first step identifies all prediction errors
 340
S. PANKAJ AND M. DUA

Table 1. Categories and their reference.

Confusion-Diffusion Cellular Automata Compression Image Watermarking
Only – Chaos map based based Neural Network based DNA based based based based Miscellaneo-us
(Ibrahim et al., 2020) (Jeevitha & Amutha (Selvi et al., 2021) (Ravichandran et al., 2021) (Choi et al., 2019) (Hajjaji et al., (Thanki & Kothari, (Hafsa et al., 2021)
Prabha, 2020) 2019) 2021)
(OISHI & Inoue, 1982) (Masood et al., 2021) (Zhou et al., 2020) (Guesmi & Farah, 2021)vv (Ismail et al., 2018) (Raja, 2019) (X. B. Han et al., 2020) (M. Kumar et al., 2021)
(Sangavi & Thangavel, 2020) (Kamal et al., 2021) (Lakshmi et al., 2020) (El-Shafai et al., 2021) (Wang et al., (Joshua C Dagadu & (Deb & Bhuyan, 2021)
2016) Li, 2018)
(S. Kumar et al., 2019) (Hua et al., 2018) (B. Han et al., 2020) (Barik & Changder, 2021) (Shen et al., 2021)
(Cariolaro et al., 2002) (Y. Chen et al., 2020) (Elamir et al., 2021) (Shafique et al., 2021)
(Ke et al., 2019) (Gafsi et al., 2020) (Akkasaligar & Biradar, 2020) (Yin & Li, 2020)
(Ismail et al., 2018) (Chai et al., 2019) (Dagadu et al., 2019) (H. Liu et al., 2019)
(Fu et al., 2018) (Moafimadani et al., (Stalin et al., 2019) (Shankar et al., 2018)
2019)
(X. Chen & Hu, 2017) (J. Liu et al., 2018) (Ravichandran et al., 2017) (Nematzadeh et al., 2018)
(Chandrasekaran & (Parvees et al., 2018) (Parameshachari et al., 2017) (Singh et al., 2018)
Thiruvengadam, 2017)
(Reddy & Fathima, 2016) (Akkasaligar & Biradar, 2016) (Zhang et al., 2017)
(Prabhavathi et al., 2018)
(Cao et al., 2017)
(Parvees et al., 2017)
(Saraswathi & Venkatesulu, 2017)
(Dagadu et al., 2017)
(Dai et al., 2016)
(Pareek & Patidar, 2016)
(Ahmad & Hwang, 2016)
(Ravichandran et al., 2016)
(Dagadu et al., 2016)
(El Assad & Farajallah, 2016)
(Tang et al., 2016)
 INFORMATION SECURITY JOURNAL: A GLOBAL PERSPECTIVE 341

map. This value is XORed. The process is repeated

until all pixel values have been processed. The
correlation coefficients show that the desired out­
come can be obtained after two whole iterations.
Chen et al. (X. Chen & Hu, 2017) proposed an
encryption scheme that uses multiple chaotic maps.
Here, the logistic sine chaos maps scrambled the
source image. Pixel diffusion happened by taking
XOR of a particular pixel with its adjacent pixels.
The author defined four types of diffusion meth­
ods. The selection of the diffusion method depends
Figure 7. Percentage of studies in various categories. on the key value used. Then, the division of the
scrambled image into two by two blocks took place.
All sub-blocks were encrypted using a hyperchaotic
and stores them in a binary map of the error loca­ system. Again, depending on the pixel positions,
tions. Here, the image is pre-processed according there are four types of encryption methods, and
to the error location map to protect the clarity of they chose type-1 encryption. The NPCR and
the picture. In the second step, the image is UACI values obtained were 99.59% and 33.42%.
encrypted using a chaotic generator based on the In 2017, Chandrasekaran et al. (Chandrasekaran
piecewise linear chaotic graph. Using the key, each & Thiruvengadam, 2017) proposed the idea of
pixel gets encrypted. And lastly, a high capacity encryption using the number-theoretic approach
reversible data hiding method (CPHCRDH) is with Henon map. The two-dimensional key array
used to correct these prediction errors, and we get was generated based on modular exponentiation.
the ciphered image. Experimental results validate In their process, the key matrix was permutated by
the working of proposed algorithms. a 2-D Henon map to create keys for each pixel of
The authors (Ismail et al., 2018) have proposed the DICOM image. Furthermore, the bitwise XOR
a Double-humped (DH) Logistic map-based of the image with permutated key matrix resulted
encryption scheme. DH logistic map is used for in encryption. The histogram analysis, correlation
pseudo-random number key generation (PRNG). coefficients, and other evaluating parameters
There is a visible similarity between the bifurcation proved that this algorithm was decent enough.
diagram of the DH logistic map and the classical In (Reddy & Fathima, 2016), the authors had used
logistic map, but the difference lies in the region a combination of stenography and encryption. The
where r = 7. The existence of more than one chaotic plain image is encrypted first by the Henon map and
region is visible. The paper has analyzed the posi­ later by the Chebyshev map. Histogram analysis
tive and negative sides of the bifurcation diagram shows that this algorithm was not the perfect one
separately. For more in-depth information, kindly and hence required many improvements.
refer to other articles.
Shan et al. (Fu et al., 2018) propped a scheme
4.2. Confusion - diffusion based encryption
that employs multiple 1-D chaotic maps for per­
algorithm
forming encryption. They aim to avoid the draw­
back caused due to the usage of invertible area- In (Jeevitha & Amutha Prabha, 2020), Jeevitha et al.
preserving chaotic maps by introducing a pixel introduced a discrete wavelet transform (DWT)
swapping method. At first, the image is trans­ based block scrambling method. This method
formed into a 1-D array. Scrambling replaces the used edge maps that were extracted from the
pixel value and takes the chaotic sequence as an source image. It comprises DWT-plane generation,
index reference. This process is followed by sub­ edge map sequences generation, and DWT-level
stitution; here, a chaotic sequence is randomly scrambling for weakening pixel correlation. In the
selected from a group of three chaotic 1-D maps; first stage, the image is decomposed into DWT
logistic-sine map, logistic-tent map, and tent-sine planes. In the second stage, n edge maps are
342 S. PANKAJ AND M. DUA

produced from chaotic maps from the same or rounds, having two blocks each. The initial input
different thresholds equal to DWT planes pre­ is a plain image inserted with random data, and the
viously generated. For this purpose, the double- output is a ciphered image. Two pseudo rows and
sine map is used. Lastly, XOR is performed columns are appended to the source image for
between edge maps and DWT planes. After this, random data insertion. Pixel adaptive diffusion is
the position of XORed-DWT planes are scrambled done in Zigzag order. Chen et al. (Y. Chen et al.,
using n edge maps that used the Sine map, and 2020) proposed an improvement on the above
lastly, inverse DWT is performed to obtain the scheme. They found out that it suffered for good
ciphered image. The keyspace is around2150 . performance under bad randomness conditions.
In (Masood et al., 2021), Masood et al. intro­ To improve it, non-linear operations are per­
duced a scheme based on Henon’s chaotic map, formed with the slightly modified algorithm.
Brownian motion, and Chen’s Chaotic system Logistic sine system (LSS) generates the initial con­
(CCS) to encrypt medical images. At first, the ditions and scrambling matrix, followed by two
source image is resized in 512 × 512 pixels and rounds of high-speed scrambling and adaptive per­
then divided into an equal number of blocks. First mutation. LSS is a pseudo-random number gen­
shuffling using 2-D HCM in each block is done to erator that takes up a secret key. Non-linear
achieve inter-block shuffling. Then, inter-cell shuf­ operations are performed between the scrambling
fling for each block occurs, followed by again shuf­ and diffusion state twice. The resulting image is the
fling of the blocks. 3-D Brownian motion gets cipher image. The security analysis and random­
initiated and iterated 1,536,256 times, defining the ness testing validate the improvements done.
number of particles and the position of particles In (Gafsi et al., 2020), Gafsi et al. have intro­
giving out a three-direction sequence. One duced a three-step encryption algorithm. The
sequence is randomly selected. After the multipli­ PRNG uses a Dynamical state generator, Henon’s
cation operation, absolute and round functions are map, 2-D logistic map, and Baker’s map. PRNG
applied on that sequence. CCS is used for extra also gives out a sequence of independent numbers
security along with bitwise XOR operation on the called PRNS. In the first step, the PRNG-generated
chosen series and earlier obtained image. Thus, the matrix is used random permutation. Next, using
output is the ciphered image. some internal condition, the pixel bits are again
In (Kamal et al., 2021), Kamal et al. utilized the permutated. From the PRNG, 2 S-Boxes were gen­
Confusion and Diffusion concept. At first, the erated. In the second step, substitution is per­
image is split up into blocks. Then, these blocks formed using both S-boxes and the previous
are scrambled using Zigzag pattern scrambling, permutated matrix. Lastly, the pixels XORed with
rotation, and random permutation. Next, the key PRNS to give the intermediate encrypted image. If
generated by the logistic map diffuses the the number of rounds condition is satisfied, this
scrambled image. The Zigzag pattern for scram­ becomes the final encrypted image; it also goes
bling is applied on both undivided sub-blocks and back to the initial permutation condition.
blocks. A random vector is produced, which helps In (Chai et al., 2019), Chai et al. proposed an
in random permutation between blocks. The encryption scheme with three steps. At first, initial
resulting matrix after random permutation is con­ values are computed using the SHA-256 hash value of
verted into the 1-D matrix and XORed with the the source image. These initial values are substituted
key, changed back to the 2-D matrix to obtain the in the 4-D Memristive chaotic system. Secondly, two
ciphered image. Decryption is the inverse of key matrices and one index matrix are constructed
encryption. We observed that the PSNR value is through the chaotic sequences generated. This index
not close to 8; this can be improved. matrix is used with permutation-based on plain
Hua et al. (Hua et al., 2018) formulates a high- image and Latin square (PPILS). PPILS simulta­
speed scrambling and pixel adaptive diffusion neously shuffles rows and columns. The resulting
approach for encryption. The structure is straight­ matrix is the shuffled matrix. This matrix is then
forward. There are two rounds of scrambling and rotated 180 degrees clockwise and sent for the next
diffusion. The secret key is distributed on two step. In the third step, bi-directional adaptive
 INFORMATION SECURITY JOURNAL: A GLOBAL PERSPECTIVE 343

diffusion is performed. Here bi-direction means con­ image. The ECEM type-3 and type-4 generate the
struction of forward diffusion path (FDP) followed by swapping sequence. Swapping sequences exchange
reverse diffusion path (RDP). For FDP, key matrix the pixel values, and 16 such sequences are
one is utilized, and for RDP, key matrix two. These required, 8 for each block. ECEM type-2 generates
steps result in the encrypted image. For decryption, diffusion sequence. This sequence is converted into
inverse PPILS is applied along with inverse FDP and an integer values sequence, and eight such
inverse RDP. This algorithm works well for binary sequences are required. First, Permutation is per­
images too. When tested against Gaussian noise, the formed, followed by the first swap. Then, diffusion
results shown are good. is performed, followed by a second swap.
In (Moafimadani et al., 2019), the authors have
proposed to encrypt an image using high-speed 4.3. Neural network-based encryption algorithm
permutation and adaptive diffusion. They have
Selvi et al. (Selvi et al., 2021) proposed a scheme
used a Chen-based hyperchaotic system. On the
Adaptive sigma filterized synorr certificate-less
plain image, SHA-256 is applied to get the initial
sign-cryptive Levenshtein entropy coding-based
values. A set of seed values are put in the chaotic
deep neural learning (ASFSCSLECDNL) encryp­
system. Four different vectors are obtained from
tion technique for medical image. Sigma filter,
the chaotic system on which the circular shift is
also called smoothening filter, denoises the image,
performed to obtain the Scrambling effect. The key
and synorr certificate-less sign-cryption takes care
matrix obtained from the chaos system aids the
of signature generation and encryption.
diffusion. Adaptive diffusion consists of several
Levenshtein entropy encoding does the compres­
mathematical operations like bitwise XOR, bitwise
sion. The working of the proposed algorithm is
circular shift. The decryption process is the reverse
such that it first pre-processes, then encrypts, and
of the encryption process. The proposed algorithm
then compresses the image. First, the image is
shows resistance against Occlude attack.
decomposed into vertical and horizontal segments,
In (J. Liu et al., 2018), the authors have presented
and then pixels are arranged in a kernel to find the
a simple encryption scheme that uses a hyperbolic
center pixel and deviation. Based on this noise,
sine chaotic system along with a confusion-diffusion
pixels are removed. The pre-processing stage ends
system. At first, the average pixel value is obtained
here. For encryption and signature generation, ses­
after pre-processing the image and is sent to the
sion key, private key extraction, and private key
chaotic system. The pre-processed image is sent for
construction are done. The image is encrypted
diffusion, where the chaotic system aids it. Diffusion
using these public-private keys, generating
directly results from XOR between the chaotic
a digital signature. Code words are set using the
sequence and the pre-processed image. After diffu­
neural network, their lengths are measured, and
sion, Column permutation followed by Row permu­
weight and probability are computed. The output
tation takes place. The diffusion and permutation
of NN gives the compressed image. Reversal of
blocks together make up the De-correlation opera­ these steps gives the decrypted original image.
tion block. The decryption is just the inverse of In (Zhou et al., 2020), Zhou et al. proposed an
encryption. The NPCR and UACI values obtained encryption algorithm based on Game theory, opti­
in the paper are good. Nevertheless, this algorithm mized Region of interest (ROI) parameters, and
suffers from the trade-off between complexity and Hidden ROI position. Quantum cell Neural network
speed. The speed is good, but it can be cracked as the (QCNN) generates random sequence utilized in
algorithm is less complex. scrambling and diffusion of ROI. QCNN is stable
In (Parvees et al., 2018) the authors have pro­ with four Lyapunov exponents making it
posed a four-stage scheme for DICOM images. The a hyperchaos system. At first, the image is segmented
ECEM type-1 (Parvees et al., 2016) generates two using the Game Optimization model. Then, ROI is
sequences that are sorted and together make the selected according to the optimal image segmentation
permutation sequence. Their proposed scheme threshold. Next, ROI is changed from the spatial
takes up eight such sequences to scramble the domain to the transform domain using the
344 S. PANKAJ AND M. DUA

2-D lifting wavelet transform. Scrambling and diffu­ to get the encrypted image. The decryption is the
sion operations are performed using the vectors inverse of the encryption.
obtained from wavelet transform (hyperchaos sys­
tem). Then, it is converted into pixel-level image for­
4.4. DNA-based encryption algorithm
mat to achieve lossless encryption/decryption. Finally,
the entire ROI is encrypted later combined with DNA is a two-stranded, double helix structure
Region not of interest to obtain the ciphered image. genetic material that provides the base for our
Lakshmi et al. (B. Han et al., 2020) used Hopfield existence. It is composed of four nucleotide bases,
Neural network (HNN) in conjunction with back­ namely adenine (A), thymine (T), guanine (G), and
propagation neural network (BPN) for medical cytosine (C). The pairing bases A and T, G, and
image security over cloud storage. There is a total C are complementary to each other. Twenty-four
of four stages. In the first stage, the image-specific kinds of DNA rules can be formed using the four
key is generated using the BPN. The BPN uses an bases. We only consider 8, given in Table 2. For
image database on which it gets trained after fea­ DNA encoding, a number is converted into an
ture extraction. BPN analyses the given source 8-bit binary form and then encoded according to
image on the basis of its previous training and the rules ranging from 1 to 8. For instance, we have
generates the key. Suppose the number of epochs 198; when converted in binary format, it gives
is increased, then the MSE property value 11,000,110. When encoded by rule 1, it gives
decreases. Next, in the second phase, random chao­ TAGC. The benefits of using DNA include a large
tic sequences are generated using HNN and pre­ degree of parallelism resulting in high computation
viously generated image-specific key. This speed, low power consumption, and large informa­
sequence is used in the third phase for adaptive tion transmission capacity.
confusion. Third and fourth stages are about adap­ In (Ravichandran et al., 2021), Ravinchadran
tive confusion and adaptive diffusion. The authors et al. proposed an algorithm that takes the benefit
have kept the fifth stage for connection establish­ of hybrid DNA blended with integer wavelet
ment and upload on cloud services. Hence, this transform (IWT) for securing medical images.
completes the whole cycle of encrypted image gen­ The algorithm works in two phases, namely,
eration and cloud storage. Histogram analysis shuffling and diffusion phases. In the first
shows a decent enough uniform pixel distribution. phase, the pixel shuffling of the whole block
In (Ravichandran et al., 2021), Han et al. pro­ takes place based on the key sequences obtained
posed an encryption scheme using a logistic map from chaotic maps. They have used 1-D logistic
and Hermite neural network. The logistic map and sine map for this purpose. The shuffled
generates the sample training chaotic sequence. image is transformed into an IWD coefficient
The next step is to establish the structure and and divided into four sub-bands. Only the LL
training indicators for Hermite NN. Then, it is band is considered. Again using the sequences
trained on the sample chaotic sequence. Source generated by the previous two chaos maps, row,
image size and Two initial values are fed into and column shuffling is performed on the LL
trained Hermite NN, which gives out the two chao­ sub-band. The last operation of stage one is cir­
tic sequences further used for encryption. They cular row and column-wise shifting. This is done
took the sum of these sequences. For the encryp­ to provide extra security. In the second stage,
tion, Round (Sum *256) is multiplied by another diffusion is performed with the help of DNA
coefficient to give a sum of 1. The sum is performed encoding and DNA XOR operations. Again the
image is DNA decoded, and after inverse IWT,
cipher image is produced. The decryption process
Table 2. DNA encoding rules. is the reverse of encryption. The keyspace turns
1 2 3 4 5 6 7 8 out to be 10140 which is fairly good.
A 00 00 01 01 10 10 11 11 Guesmi et al. (Guesmi & Farah, 2021) proposed
T 11 11 10 10 01 01 00 00
G 01 10 00 11 11 11 01 10 an algorithm based on hybrid DNA masking, using
C 10 01 11 00 00 00 10 01 SHA-2 and a new hybrid chaotic map to improve
 INFORMATION SECURITY JOURNAL: A GLOBAL PERSPECTIVE 345

IE. The new chaotic map depends on sine, logistic phase, the Tent map uses the aforementioned
and tent maps. The algorithm starts with generat­ initial conditions and control parameters to per­
ing keys through SHA-2 from the source image. form XOR operations for employing the concept of
This key is fed into two instances of the hybrid map confusion. Thus, ciphered image is obtained. There
and DNA encoding. DNA encoding is performed is one Tent map for each block, and the decryption
on the image to convert it to an image vector. It is process is the reverse of the encryption process.
further divided into eight sub-vectors. The gener­ The complete set of keys, including RSA-
ates S-box is applied to each sub-vector. protected seeds, are encrypted by AES to increase
The second instance of the hybrid map is iterated the security during the key exchange between
row*col times of the image. The pixels are shuffled Sender and Receiver.
using the values of the random permutation vector. Elmair et al. (Elamir et al., 2021) focus on hiding
This whole process is carried out 10 times. It is the information in the least significant bit of image
observed that the obtained histograms were not pixels of a medical image. LSB is used here because
very uniform, and the scope of improvement is it will not affect the look of the image much if data
there. is hidden inside it as it falls farthest to the right.
In (El-Shafai et al., 2021), Shafai et al. pro­ A key is generated from six-stage chaotic maps and
posed a scheme using a Logistic chaos map, is used for image compression. These six maps are
piecewise linear chaotic map (PWLCM), MD5, the Chebyshev map, Gauss map, Henon map,
and DNA encoding. PWLCM is used for key Logistic map, Tent map, and Piecewise map.
image generation because it has two stages of These maps are applied in sequence over the partial
confusion and diffusion. In the confusion step, ciphered image. For encryption of this image, DNA
pixels are adjusted with the XOR technique. Pixel encoding is done to get the final ciphered image.
positions are changed according to the random While using the maps, sequence matters; if execu­
permutation key in the diffusion step. At first, tion is not done in order, results will vary.
PWLCM uses the logistic map to generate the In (Akkasaligar & Biradar, 2020), Akkasaligar
key image. Row-wise DNA encoding is done et al. came up with the idea of selective encryption
based on the rule selected by the PWLCM key with DNA encoding. They have used a pixel select
on both key and source image. DNA-XOR is method that divides the image into matrices M1
performed between them to obtain an intermedi­ and M2. M1 keeps the selected pixel while M2
ate image. Then, row rotation followed by col­ stores the remaining pixels. Both the matrices are
umn rotation is done. Later, column-wise DNA DNA encoded. The M1 matrix is pixel shuffled
encoding followed by DNA XOR is performed to using the dual hyperchaos map, and the resulting
give a cipher image. matrix, along with the M2 matrix, is diffused
Barik et al. (Barik & Changder, 2021) formulated together by DNA XOR operation. DNA decoding
a scheme that works in two phases. This scheme is performed on the resulting matrix, and thus we
uses 20 Amino acid-based codons. Codons are used get the ciphered image. NPCR and UACI values are
for DNA code substitution. A particular DNA code 99.87% and 33.29%, respectively. The proposed
can have more than one alternate codon. In the algorithm is not complex and performs well.
case of the presence of multiple alternate codons, A mixture of two chaotic systems, the Bernoulli
Linear Congruential Generator is used. For addi­ shift map, and the ZigZag map, is combined with
tional security, the seeds are encrypted by RSA. In DNA encoding in (Dagadu et al., 2019) by Dagadu
phase 1, the image gets substituted by the DNA et al. The whole process works in three stages. At
code and then into an Alternate codon sequence. first, the MD-5 hash function is applied to the
Then, it gets divided into blocks. A logistic map is image. This hash value is used with an ASCII string
used to generate unique control parameters and input to generate the initial conditions and control
initial conditions for each block. The seeds are parameters for the two chaotic systems. Both the
random ASCII characters generating a Pseudo- chaotic systems generate two key matrices.
random number generator. Then, each block is Secondly, a Logistic map is iterated to produce
circular shifted using this PRNG. In the second the sequences to be used in a row by row diffusion
346 S. PANKAJ AND M. DUA

between the two matrices and the plain image They initialized the chaotic logistic map and
matrix. For this, DNA XOR is used. Again the created the Sudoku matrix from it. The source
output of the logistic map decides which DNA image was DNA encoded by selecting
encoding rule will be used. The DNA-XORed a particular encoding rule. Dual DNA addition
matrix is then DNA encoded using the previously refers to performing DNA XOR twice. So, the
decided rule. The final product is the cipher image. DNA encoded image gets XORed twice, and the
The decryption method is symmetric and inverse output was a partially encrypted image.
of the encryption process in nature. Histogram Akkasaligar et al. (Akkasaligar & Biradar, 2016)
analysis produces good results. NPCR and UACI combined Chao’s theory and DNA encoding to
values also show resistance to differential attacks. produce a new encryption algorithm. They had
Stalin et al. (Stalin et al., 2019) proposed the first encoded the image matrix into two DNA-
NL4DLM_DNA technique, a 4-D Logistic map coded matrices. Next, based on pixel value, chaotic
composition followed by DNA encoding. Plain sequences were created utilizing the DNA matrix.
image is divided into 16 blocks, and for each If the pixel value was odd then, Chen’s hyperchao­
block, a chaotic sequence is generated using tic map was else Lorenz chaotic map was used.
a non-linear 4-D logistic map. These sequences Pixel positions of both DNA encoded matrices are
are converted to binary format and then are scrambled using the generated sequences. Finally,
DNA encoded using rule 3. Parallelly, for each both the matrices were added using the DNA ADD
block, pixel substitution is performed to generate operation to get the encrypted image.
a binary series. This sequence is also DNA In (M. Kumar et al., 2016), Kumar et al. pro­
encoded using rule 1. DNA subtraction is per­ posed to encrypt the image using Elliptical curve
formed between the two DNA sequences to gen­ Diffie-Hellman cryptography (ECDHC) and DNA
erate a new one. This new sequence is again encoding. The image is split into R, G, and B color
DNA encoded using rule 1. Bitwise XOR is per­ channels. DNA encoding, DNA addition followed
formed between this sequence and the binary by Shifting within layers were performed. After
logistic map sequence. This gives the final that, ECDHC is applied to get the final encrypted
encrypted image. Keyspace is estimated at image. Contrast differences can be noticed in the
around 1060 original and decrypted Lena image. Hence, this is
In 2017, Ravichandran et al. (Ravichandran not a preferable algorithm.
et al., 2017) discussed an encryption algorithm
that had three phases, permutation, encoding,
4.5. Cellular automata based encryption algorithm
and diffusion. The image was decomposed into
planes. The coupled logistic sine (CLS) map Choi et al. (Choi et al., 2019) proposes a scheme of
generates the sequence which permutates the color medical image encryption comprising of
split planes. Then begins the encoding phase. 3-D chaotic Cat Map and a pseudo-random num­
CLS provides input to Coupled Logistic tent ber generator, NCA. NCA is a group of cellular
(CLT) system, which provides a chaotic automata obtained by combining two non-linear
sequence for DNA encodings. DNA sequence cellular automata (CA) and a non-linear maximum
encoder, DNA sequence compliment, DNA length CA (MLCA). From the key, NCA and
sequence adder/subtractor, and DNA sequence 3-D cat maps are initialized. The plain image is
decoder were placed one after the other to com­ divided into three channels, RGB. These channels
plete the DNA encoding phase. Again, CLS pro­ are then XORed with the output of NCA. The
vided chaotic sequence input to diffusion phase result is then combined, and a 3-D Cat map is
where XOR operation took place and outputs applied “n” number of times to the shuffled
the final image. The estimated keyspace was image. This shuffled image is the encrypted
nearly 10168 image. The algorithm can be improved and
In (Parameshachari et al., 2017), the authors extended to cover the grayscale images too.
had proposed an algorithm of dual DNA addi­ Jeong et al. (Ismail et al., 2018) have proposed
tion to getting the partially encrypted image. a color encryption scheme that uses a 2-D chaotic
 INFORMATION SECURITY JOURNAL: A GLOBAL PERSPECTIVE 347

map and C-MLCA. Cellular automata (CA) is Wang et al. (Wang et al., 2016) proposed
a discrete-time dynamic system. MLCA is a scheme where compression and compression
a maximum length CA that has a feature that were simultaneously happening based on tensor
minimum polynomial is the primitive polyno­ compressed sensing with a 3-D Lorenz system.
mial. CMLCA is a combination of XOR and Three sensing matrices are initialized. The images
XNOR logic. At first, the Wolfram rules are and sensing matrices go through ALT-TCS (tensor
used to create a transition matrix. Then, the PN compress sensing), which outputs optimized
sequence is generated using MLCA. Next, matrices and encrypted plus compressed tensor.
CMLCA is generated using XOR with comple­ To get decrypted, these obtained matrices need to
mented vector. Three CMLCA images are created pass through the 3-D Lorentz system.
using the sequences generated, and each is
XORed with the three color channel matrices;
R, G, and B. Lastly, a 2-D chaotic transform is 4.7. Image watermarking based encryption
applied to get the ciphered image. They have algorithm
tested this algorithm on 100 different colored
Thanki et al. (Thanki & Kothari, 2021) proposed
images.
a scheme in which they insert the secret informa­
tion into a cover image using hybrid finite ridgelet
4.6. Compression based encryption algorithm transform (FRT) and singular value decomposition
(SVD) for authentication and identification. This
Hajjaji et al. (Hajjaji et al., 2019) proposed
combined image is then encrypted using the
a compression-based encryption scheme that uses
Arnold Scrambling method. FRT is preferred here
a neural network and piecewise linear chaotic map
because it decomposes the cover image into an
(PWLCM). The image is divided into blocks. The
image of its double size. This extra size availability
image is compressed using the Artificial NN
makes insertion of secret information easy and
(ANN). The decision on the number of training
payload management easy. The process starts with
iterations, layers and other parameters becomes
crucial. The ANN used in the algorithm is the application of FRT and then SVD on the cover
Multilayer Perceptron (MLP). Then, Arnold’s cat image. Only SVD is performed on the watermarked
map is a 2-D chaotic map. PWLCM is used to image. Then insertion of a watermarked image in
modify the values of the hidden neural network the cover image takes place. The process of SVD
layers. First, the image blocks go through the sig­ and FRT is reversed to get the mixed watermarked
moid function and into the input layer of MLP. image. This image gets encrypted by Arnold
PWLCM affects the hidden layer and Arnold cat Scrambling using a secret key. The resulting
map shuffles the weight matrix. The output is the image is the watermarked encrypted image ready
compressed ciphered image. The demonstrated to be sent. The scheme performs well when tested
algorithm in the paper works with 36 input layers, for noise resistance through salt and pepper and
18 hidden layers, and 36 output layers. Gaussian noise test.
In (Raja, 2019), Raja proposes a compression- Han et al. (X. B. Han et al., 2020) have simulated
encryption scheme. There is a total of three stages the idea of medical image watermarking blended
outlined. On the source image, a Multiscale trans­ with a biometric authentication system. Firstly,
formation is done. These include Wavelet, DWT is applied to the cover image to get the four
Bandelet, Curvelet, Contourlet, and Ridgelet trans­ bands. Consider only three bands, LH, HL, and
forms described in the referred paper. Next, RSA HH. Each pixel of these bands goes through the
public cryptography is performed on the trans­ set of some mathematical operations. These math­
formed image. The encrypted image then goes ematical operations utilize the watermarked medi­
through Encoding techniques such as EWZ, cal image, the fingerprint image using wavelet
SPIHT, ASWDR, and WDR. The author has eval­ filters. In the end, inverse IWT is performed to
uated performance parameters values on each obtain the ciphered image. In the decryption
encoding and transformation technique. stage, the image is again divided into four sub-
348 S. PANKAJ AND M. DUA

bands. From the HL sub-band, the fingerprint The scope of more improvement is seen if
image is extracted. The watermarked image is con­ more permutation methods are introduced.
structed from LH and HH sub-band after the fin­ Kumar et al. (M. Kumar et al., 2021) introduces
gerprint match is done. There are 37 different an encryption scheme using 1-D logistic map with
wavelet filters present, and the level of decomposi­ pseudo-random numbers aiming to avoid chosen
tion is decided by the user by the “n” level. Hence, and known plain-text attacks. For this, they have
the keyspace becomes 37n . inserted two pseudo columns and two pseudo
Dagadu et al. (Joshua C Dagadu & Li, 2018) rows. Four matrices of same size are obtained after
proposed an IWT-LSB watermarking and random applying logistic map to each pixel of source image
permutation-based chaos encryption. They have and are not to be kept secret. Each element is multi­
used a chaotic Tent map. On the plain image, plied by 1010 and taken floor to get a integer num­
IWT is performed, which fived four sub-bands. ber. Modulo 256 is taken of these integers so that
Context is the information that characterizes the they lie in pixel value range. The earlier mentioned
entity. We feed this information in CV and the EPR pseudo rows and columns are filled with these
value in the CH sub-band. Next, inverse IWT is values using mathematical equations. Another
performed to get the watermarked image. The ran­ index matrix is generated which is used for pixel
dom permutation is performed to accomplish the shuffling. Then, after shuffling, XOR is performed.
encryption. The result is XORed with a Tent map This process is repeated three times. In the last step,
key sequence to get the final ciphered image. They another matrix is generated using the mixture old
have also proposed a secure architecture suitable and new keys and multiplying each element with
for sharing medical images between various 1010 and then taking mod 256. Last generated index
devices, including servers and the internet. matrix is XORed to get final ciphered image matrix.
Deb et al. (Deb & Bhuyan, 2021) formulated
a scheme using a special nonlinear filter function-
4.8. Miscellaneous
based feedback shift register (LFSR) based on
In (Hafsa et al., 2021), Hafsa et al. proposed an PRNG. They have designed an 8-variable Boolean
encryption scheme, which is implemented on function in coordination with LFSR to produce
Field Programmable Gate Arrays (FPGA) for a good quality pseudo-random number. This is dif­
medical images. They have used chaos-based ficult as by increasing the non-linearity, the func­
Pseudorandom number generator (PRNG) and tion’s balance property may get affected. Firstly, the
Modified Advanced Encryption Scheme image is reshaped, and a circular shift is performed.
(MAES). The image is decomposed in blocks Later, image is randomized using Logistic-Tent
of 128 bits. There is a total of three steps. chaotic map and then scrambled by the Arnold
First, the PRNG uses a 2-D logistic map and transformation method. Next, it is XORed with the
Henon system for key generation. Secondly, sequence generated by PRNG to get the ciphered
four rounds of encryption use four S-boxes for image. The scope of improvement lies in developing
substitution. Here to avoid the drawbacks of a better performing Boolean function.
normal AES, a 4-D chaotic system is mixed to In (Shen et al., 2021), Shen et al. introduced
get the modified version. Thirdly, instead of a selective encryption algorithm based on fast and
shift-row and mix-column operation, a random robust Fuzzy C-means clustering (FRFCM) and
permutation method is implemented. There are Facial biometric for medical images. The first step is
two predefined permutation methods; method to get the Region of Interest (ROI), i.e., the area that
one is used if the position of the block is odd; contains the primary information using FRFCM. The
else, method two is used. And lastly, each byte number of clusters, fuzzification parameters, filtering
uses XOR to round the key, and a subkey is window size, and minimal error threshold are
obtained from PRNG. This whole process runs decided. Using these values, the computation of his­
four times, and after the collection of all the togram the ROI is obtained. The coordinates are
blocks, the final ciphered image is obtained. saved. This ROI can be of any shape, but it is reshaped
 INFORMATION SECURITY JOURNAL: A GLOBAL PERSPECTIVE 349

into a rectangle in the next step. For encryption, First, the calculation of the Optimal Fitness value
generation of chaotic face phase mark (CFaPM) is (OF1 and OF2) is set after analyzing the image.
done. CFaPM is a combination of chaotic Standard Utilizing the OF1, the image goes under genetic
map and facial biometric. The rectangular ROI is selection and crossing. Next, scrambling is per­
encrypted in the Gyrator transform domain using formed based on GSAPSO. OF2 is calculated on
the phase mask. The earlier saved coordinates are this image, and if OF2 is less than OF1, the whole
mapped again, giving the final selective-encrypted process of genetic selection and crossing is carried
image. Facial biometric guarantees uniqueness in out again. If not, then the process of mutation is
the generation of CFaPM. Selective encryption is carried out. Mutation takes chaotic sequence gen­
a good idea, but the only drawback is that it will erated by iteration Henon map. The mutation is
give the attacker a fair idea about the shape of the scrambling using the chaotic sequences generated.
crucial data present in the image. Some conclusions Finally, all the mutation sequences are obtained to
can still be drawn from this selective encrypted image. obtain the encrypted image. The keyspace is sig­
Shafique et al. (Shafique et al., 2021) proposed an nificant as this algorithm employs SHA-256 for key
encryption algorithm using a cubic-logistic map, generation.
Discrete Wavelet Transform (DWT), and a bit- Liu et al. (H. Liu et al., 2019) formulated
plane extraction method. Total three cipher images a scheme for medical colored images that use
are generated. There is a total of three stages where coupled hyperchaotic system and Galois field-
in the first and last stage, the image is encrypted in arithmetic operation. Firstly, the generation of
the spatial domain, and in the middle stage, it is two state variables takes place using the Runge-
encrypted in the frequency domain. Pixel values Kutta method by iterating it 300 times. Next,
are directly manipulated by mathematical opera­ a keystream for the first round is generated by
tions such as cyclic shift, substitution, permutation using a specific set of equations given in the
in the spatial domain. In contrast, one has to con­ paper. Earlier, the image was divided into three
vert the pixel values into the different frequency color channels; R, G, and B. These color matrices
domains to manipulate them. For instance, DWT are converted into a 1-D matrix and XORed with
divides into four sub-bands. LL is considered the previously generated keystream. Next, the
because most of the plain images are present in result is permutated by rows and columns. Then,
this sub-band. The method of bit-plane extraction round sequences are generated by using Galois
divides the plain image into eight-bit planes, each field multiplication. This was the diffusion phase.
plane containing 0 and 1 as values. The MSB bit More rounds can be carried out to increase the
plane contains most of the valuable information. complexity. The output is the encrypted image.
The algorithm starts with the input image going The speed of this algorithm on a standard
through bit-plane extraction. The first instance of Windows 10 i-7 powered PC is 63.89Mbit/s. This
chaotic sequence helps in the scrambling of the last speed is considered good.
four planes. DWT, along with the second instance Shankar et al. (Shankar et al., 2018) have tried to
of the chaotic sequence, is XORed on all bit plane’s overcome the problem of smaller keyspace and low
LL band. Inverse DWT is applied. Using leftover security of the preexisting models through their
keys, the key image is generated. After mixing up algorithm. To choose the optimal public and pri­
the two images, another application of bit-plane vate keys, they have implemented the Adaptive
extraction results in three ciphered images. The Grasshopper Optimization (AGO) algorithm with
estimated keyspace is nearly around 2135 . PSNR and CC fitness function. They pointed that
In (Yin & Li, 2020), Yin et al. proposed the idea Public-Private key cryptography will get help from
of medical image encryption using a modified the selected optimal key.
quantum chaos system (QCS) and genetic simu­ Nematzadeh et al. (Nematzadeh et al., 2018)
lated annealing particle swarm optimization proposed the usage of coupled lattice and hybrid
(GSAPSO). MD5-SHA256 is implemented to pro­ genetic model for encryption. Coupled map lattices
tect the key values and seed values. These seeds and (CML) are chaotic functions that use a logistic map
key are placed in QCS to generate key streams. for sequence generation. Key generation is the first
350 S. PANKAJ AND M. DUA

phase. In the second phase, the initial population in two significant drawbacks of this algorithm. First,
the Genetic algorithm takes place by deriving ROI has to be manually selected. Furthermore, the
diverse encryption images from the source image. manually selected ROI was always rectangular. As
These are divided further into groups to evaluate these were the initial chaos-based algorithms, their
the fitness function. Next, selection and crossover problems have been solved in future studies.
happen, followed by mutation. In mutation, a fixed In the year 2016, Cao et al. (Cao et al., 2017)
percentage of the initial population is replaced by proposed a scheme using edge maps. The whole
the new ones. Then, they are sorted to check the process was divided into three parts: bit-plane decom­
improvement in entropy after 10 iterations. If there position, random sequence generation, and permuta­
is an improvement, the whole process is repeated; tion. Edge maps are mathematical equations that
otherwise, we get the ciphered image. show features and boundaries between two areas
In (Singh et al., 2018), authors have proposed an within an image. The first stage could use plane
encryption scheme based on group-based modulo decomposition methods such as Truncated
to create a circular shifting effect. A group is Fibonacci P-code Bit-planes decomposition, Binary
a mathematical structure. It has specific properties Bit-planes decomposition, etc. Then, “n” edge maps
like commutative, associative, etc. The colored were generated whose thresholds were regulated by
image is broken down into three color channels; chaotic maps such as a Double-Sine map. These edge
R, G, and B. Then, the key-based modulo group maps were XORed with Bit-planes. Next, all bit planes
operation is performed separately on each channel. were stacked together, and bit-level scrambling was
After processing every row, the value of d is incre­ performed to get the final encrypted image. The
mented by 1, and in the end, it is reset before selection of chaotic map, plane decomposition
processing the following color channel matrix. method, and type of image was flexible.
Combine all three matrices to get the ciphered In (Parvees et al., 2017), the authors have
image. This process does not perform well when improved PRNG’s like linear congruential genera­
evaluated for histogram analysis. tor (LCG) and XOR-shift generator (XSG) and
In (Zhang et al., 2017), Zhang et al. proposed combined them with a 2-D coupled logistic map
a simple algorithm that takes care of encryption, to provide better encryption. They proposed three
establishing session key and decryption for medical new types of XSG’s and LCG’s, obtained from the
images. At first, appropriate initial values are basic ones. They have utilized these new ones in
selected and put into the Chebyshev map. This their proposed encryption scheme, which com­
gives an n-dimensional vector which is later sorted prises confusion and diffusion. Kindly refer to the
to get a new sequence of vectors. Pixels are rear­ paper for in-depth information.
ranged using these vector values. Then, the session Saraswathi et al. (Saraswathi & Venkatesulu,
is established, and the image is sent to the other 2017) proposed an encryption scheme using
party. Decryption is symmetric but inverse in nat­ a stream cipher. This stream cipher was generated
ure. Taking this algorithm as a base, many from a pseudo-random binary sequence generator.
improvements are made by others. They used two large prime numbers with the PRBG
Prabhavati et al. (Prabhavathi et al., 2018) pro­ algorithm (refer to paper) to generate the key­
posed a scheme back in the year 2017 that com­ stream. This keystream is utilized in the mathema­
bines stenography with encryption. For encryption, tical operations on the plain image, thus giving us
they used confusion and diffusion. Lorentz map the output image. The simple algorithm gave good
aided the confusion, and the logistic map aided values for correlation coefficients.
the multiple levels of diffusion. In this algorithm, In 2017, Dagadu et al. (Dagadu et al., 2017)
first, the colored image is converted into a grayscale proposed a scheme; they used two chaotic systems
image. Then, the ROI is manually selected. Lorentz for key generation. They were PWLCM and
map and logistic maps are initialized and iterated Bernoulli shift maps. PWLCM system is based on
to get the chaotic sequence. The ROI goes through the MD5 hash function that was calculated after the
the confusion phase and then the diffusion phase. first round of diffusion. Two rounds of diffusion
Therefore, ciphered image is obtained. There are were performed to get the ciphered image. First,
 INFORMATION SECURITY JOURNAL: A GLOBAL PERSPECTIVE 351

the Bernoulli shift map generated the chaotic in the ciphered image. An affine transform is
sequence that produced the first key matrix. This defined in the paper referred to here.
key matrix was XORed with the plain image to In (Ravichandran et al., 2016), the authors intro­
complete level one diffusion. Similarly, level two duced chaos-based mutation and crossover pro­
diffusion was completed after XOR between the cess, which is now utilized in almost every chaos-
PWLCM key matrix and the semi-encrypted based genetic encryption algorithm. Dagadu et al.
matrix. The decryption operation is the inverse of (Dagadu et al., 2016) had introduced DWT com­
encryption. pression-based medical image encryption techni­
In the year 2016, Wang et al. (Dai et al., 2016) que. In the referred paper, they have further
proposed the encryption algorithm based on bit- discussed how an image is decomposed into sub-
plane decomposition. They have defined what a bit bands and how random permutation is performed
plane is and how each bit plane contains different using the key to get the ciphered image. Assad et al.
information about the image. The algorithm begins (El Assad & Farajallah, 2016) proposed encryption
after the decomposition of bit planes. The higher using a 2-D Cat map. A bit permutation layer
four bits are permutated using the Arnold map, follows a diffusion layer. After the diffusion layer
and the resulting matrix is XORed with Henon- was applied, the resulting matrix was changed from
Logistic map. The resulting image is the ciphered integer to binary format. The algorithm has
image. The algorithm is checked against many a PRNG that generates a key. The permutation
attacks. layer based on the 2-D Cat map is applied to the
In (Pareek & Patidar, 2016), the authors had previous binary matrix and later converted into
developed an algorithm that uses genetic algo­ integer format to give the ciphered image. The
rithms. It has a recombination process in which decryption process is just the opposite of
a new population is generated from the existing encryption.
ones. This process uses a fitness function which In (Tang et al., 2016), the input image was
is the adjacent pixel’s correlation. Other pro­ decomposed into bit-planes, and then the bit
cesses include crossover, initial population gen­ blocks were randomly swapped based on the
eration, and mutation processes. Crossover and sequence generated from the Henon map for all
mutation are carried out four times, and the the planes. XOR was carried out for diffusion on
resultant image is the encrypted block image. these planes, and the Logistic chaotic maps gener­
The algorithm paves the way for future genetic ated a matrix. The algorithm has a keyspace
encryption algorithms. of 7:3724 � 10134
In (Ahmad & Hwang, 2016), Ahmad et al. used
XOR and addition operation to remove the corre­
5. Comparative analysis
lation between the adjacent pixels. The chaotic
maps used for a key generation were PWLCM This section compares some randomly selected algo­
and tangent delay Ellipse Reflecting Cavity map rithms based on their categories and the values
system (TD-ERCS). The algorithm begins with obtained for some commonly used parameters as
the application of hash function on the source discussed in section 3.1 and 3.2. Table 3 shows
image and random ASCII character to obtain the whether the mentioned algorithms have computed
initial conditions and control parameters. They are the below-mentioned values necessary for perfor­
fed into the two chaotic maps to generate the key­ mance validation of the algorithm. As one can see,
stream. All columns and rows of the plain image many papers have not tested their method against
get permutated using sequence from TD-ERCS. some of the evaluation parameters. Figure 8 demon­
This scrambling happens for some n number of strates that the evaluation parameters Correlation
iterations, and then on the resulting matrix, Coefficient (CC), Entropy, NPCR, UACI are
PWLCM generated sequence is applied for diffu­ employed in a greater number of studies compared
sion. This whole process happens again and again to the other parameters. Table 4 shows
till all pixels are covered. Lastly, each pixel is a comparison of some algorithms on encryption
replaced with its Affine transform value, resulting on Lena 256 × 256 digital image. It is observed,
352 S. PANKAJ AND M. DUA

Table 3. Some algorithms and presence of their evaluation parameters.

Image Noise resistance Encryption Parameters Decryption Parameters
Author Input type Gaussian noise/Salt & Pepper test NPCR UACI ENTROPY CCH CCV CCD MSE PSNR
(Ravichandran et al., 2021) DICOM ✓ ✓ ✓ ✓ ✓ ✓ ✓
(Hafsa et al., 2021) Digital image ✓ ✓ ✓ ✓ ✓ ✓ ✓
(Jeevitha & Amutha Prabha, 2020) Digital image ✓ ✓ ✓ ✓ ✓ ✓
(Guesmi & Farah, 2021) Digital image ✓ ✓ ✓ ✓ ✓ ✓
(El-Shafai et al., 2021) Digital image ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
(M. Kumar et al., 2016) Digital image ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Deb et al. [106] Digital image ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
(Deb & Bhuyan, 2021) Digital image ✓ ✓ ✓ ✓
(Y. Chen et al., 2020) Digital image ✓ ✓ ✓ ✓ ✓ ✓
(Ibrahim et al., 2020) Digital image ✓ ✓ ✓ ✓ ✓ ✓
(X. B. Han et al., 2020) DICOM ✓ ✓ ✓ ✓
(S. Kumar et al., 2019) Digital image ✓ ✓ ✓ ✓ ✓ ✓
(Chai et al., 2019) Digital image ✓ ✓ ✓ ✓ ✓ ✓ ✓
(Choi et al., 2019) Digital image ✓ ✓ ✓ ✓ ✓
(Hajjaji et al., 2019) Digital image ✓ ✓ ✓ ✓
(Raja, 2019) Digital image ✓ ✓ ✓ ✓ ✓
(J. Liu et al., 2018) Digital image ✓ ✓ ✓ ✓ ✓ ✓
(Ismail et al., 2018) Digital image ✓ ✓ ✓ ✓ ✓ ✓ ✓
(Hua et al., 2018) DICOM ✓ ✓ ✓ ✓ ✓ ✓ ✓
(Joshua C Dagadu & Li, 2018) Digital image ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
(X. Chen & Hu, 2017) Digital image ✓ ✓ ✓ ✓ ✓ ✓
(Parvees et al., 2017) DICOM ✓ ✓ ✓ ✓ ✓ ✓ ✓
(Reddy & Fathima, 2016) DICOM ✓
(Tang et al., 2016) Digital image ✓ ✓ ✓ ✓ ✓

are presented in Figure 10. It can be concluded from

Figure 10 that the Logistic chaotic map has been
used by most of the encryption methods.

6. Open issues of image security and summary

Researchers and innovators are focusing on making
intelligent architectures that can aid human life and
make things easy. One such approach is to make
smart healthcare and cloud storage platforms.
Figure 8. Comparison of evaluation parameters in various However, the security of an image, whether it is
studies. a typical image or a medial image carrying crucial
information, makes the development of intelligent
architecture difficult. There is always a trade-off
that researchers often let go the computation of MSE between security and ease of implementation.
and PSNR values. Figure 9 illustrates the parameters Moreover, the trade-off between security and effi­
values obtained by various methods included in ciency exists. We always have to find a way to
Table 4. The comparison between some encryption balance this trade-off to find the optimal encryption
algorithms is presented in Table 5. The chaotic map solution. Still, there are some open issues in image
used by various research works included in Table 5 encryption schemes that need to be addressed:

Table 4. Comparative analysis of selected algorithms on image Lena 256 × 256.

Decryption
Encryption Parameters Parameters
Author NPCR UACI ENTROPY CCH CCV CCD MSE PSNR
(Hafsa et al., 2021) 99.69561 33.81015 - −0.0003 −0.006 0.00014 - -
(Guesmi & Farah, 2021) 99.604120 33.431710 7.997258 −0.025325 0.002558 0.009052 - -
(M. Kumar et al., 2016) 99.6166 33.4971 7.99935 −0.0052 −0.0049 −0.0117 0 ∞
(Deb & Bhuyan, 2021) 99.6017,99.6070, 99.5948 33.6194, 33.5987, 33.6027 7.26003, 7.55744, 7.02149 - - - - -
 INFORMATION SECURITY JOURNAL: A GLOBAL PERSPECTIVE 353

Figure 9. Parameter values obtained by various methods on image Lena 256*256.

Table 5. Some encryption algorithms comparison.

Author Chaotic map(s) used Number of Keys + Control parameters Keyspace
(Ravichandran et al., 2021) 1-D Logistic map, Sine map 10 10140
(Jeevitha & Amutha Prabha, 2020) Double Sine map, Sine map 4 2150
(Guesmi & Farah, 2021) Combination of Logistic-Sine-Tent map 4 > 2199
(Zhou et al., 2020) - 12 10192
(El-Shafai et al., 2021) 1-D Logistic map 5 1079
(M. Kumar et al., 2021) 1-D Logistic map 18 > 10135
(Deb & Bhuyan, 2021) Logistic Tent map, Arnold’s cat map 4 > 2256
(Kamal et al., 2021) Logistic map 3 1035
(Barik & Changder, 2021) Logistic map, Tent map 10 + n sub keys 4.109 x 2335
(Thanki & Kothari, 2021) Arnold’s map 1 key -
(Shen et al., 2021) Chaotic Standard map 6 + biometric 1095

Figure 10. Comparison based on chaotic map used by various

studies.
Figure 11. Dedicated work percentage for grayscale and
coloured images.
● Encryption schemes that can encrypt color
and grayscale medical images These schemes support color and grayscale image
encryption. In contrast, some of the schemes only
Many medical images generated are in grayscale encrypt colored images making them unfit for
with a black background. While studying the medical image encryption. A good scheme should
encryption schemes, we found out that some of be one that can support both grayscale and colored
those can also be applied to medical images. image encryption. Figure 11 shows the percentage
354 S. PANKAJ AND M. DUA

Table 6. Common encryption schemes and their advantage and disadvantage.

Techniques Advantages Disadvantages
Chaotic maps Large key space, robust against common known attacks
DNA encoding Efficient as capable of storing large amount of data, low Cannot tolerate chances of error
power consumption
Compression Memory efficient but time taking Not always required as availability of large storage space is cheap
nowadays
Watermarking - Required an additional cover image
Selective encryption Focuses only on vital part of the image for encryption Attacker can get the faint idea about the encrypted ROI by seeing the
shape encrypted image.
Neural Network Self-learning and adaptive Required proper training before starting to process important data
Confusion – Diffusion Simple and efficient Multiple iterations required for perfect encryption

of work dedicated to developing schemes that spe­ process an image encryption scheme on
cifically encrypt grayscale and colored images. a computer, we should try to utilize the CPU as
much as possible. This benefits us by decreasing the
● Memory efficient: required time and taking advantage of high-
powered processing CPUs almost everywhere.
Space utilization is not considered a problem To say it precisely, the upcoming encryption
nowadays because of the availability of infinite algorithms and research work shall consider how
cheap storage. However, not everyone can afford to utilize new and existing technology more. New
that. Therefore, data compression should also be algorithms should support cloud and IoT functions
taken care of. efficiently as they are our present and part of the
future. To summarize our study, Table 6 describes
● Recoverable if tampered upto certain limit: some of the commonly used techniques, their
advantages and disadvantages. Table 7 shows the
Some images are very crucial, like medical and best usage of these techniques.
military images. They lose their value even if
a tiny portion of them gets disturbed. Therefore,
7. Future scope and conclusion
the data quality should be handled very strictly.
A good algorithm should be resistant to noise, This paper has reviewed several image encryption
and no modification should be done during techniques to give the reader a rundown of what
encryption or decryption. The algorithm should has already been implemented. It is a fact that the
also be able to recover the image if it has meddled. requirement for a better and secure image encryp­
tion technique can never get fulfilled as, along with
● Algorithms can take advantages of current improvement in these techniques, threats are get­
technology ting stronger too. For research purposes, many
improvements can be made to cover all the open
Parallel processing is the trend today. Many algo­ issues in image security. Proposing a fully secure
rithms studied works sequentially. When we methodology is not an easy task, but learning from

Table 7. Common encryption schemes and their general best usage.

Techniques Best uses
Chaotic maps Good for Medical image encryption as it has lesser blank windows.
DNA encoding Good for big size images that are not very critical.
Compression Good for big size images.
Watermarking Good for images that focus on being Authenticated.
Selective encryption Good for Medical images as the black part don’t need encryption.
Neural Network Suitable any number of images having similar pattern.
Confusion – Diffusion Can be added in as a part of encryption scheme to any existing or new encryption scheme.
 INFORMATION SECURITY JOURNAL: A GLOBAL PERSPECTIVE 355

already proposed approaches becomes a little less Computation, 45(172), 393–403. https://doi.org/10.1090/
complicated. S0025-5718-1985-0804931-X
We found out that even small improvements in the Cao, W., Zhou, Y., Chen, C. L. P., & Xia, L. (2017). Medical
image encryption using edge maps. Signal Processing, 132,
existing methods can prove to be very effective. Every
96–109. https://doi.org/10.1016/j.sigpro.2016.10.003
technique has its own advantages and disadvantages Cariolaro, G., Erseghe, T., & Kraniauskas, P. (2002). The
as summarized in Table 6. For instance, compression fractional discrete cosine transform. IEEE Transactions on
saves memory, which can be used for transferring Signal Processing, 50(4), 902–911. https://doi.org/10.1109/
large media over a channel. In contrast, chaos takes 78.992138
up space and memory but provides higher random­ Chai, X., Zhang, J., Gan, Z., & Zhang, Y. (2019). Medical
image encryption algorithm based on Latin square and
ness to a cryptosystem making it more robust. By
memristive chaotic system. Multimedia Tools and
keeping in mind all the advantages and disadvantages, Applications, 78(24), 35419–35453. https://doi.org/10.
we must use the technique which is most suitable for 1007/s11042-019-08168-x
our purpose. To extend our work further, we wish to Chandrasekaran, J., & Thiruvengadam, S. J. (2017). A hybrid
analyze more research works and learn from them to chaotic and number theoretic approach for securing
come up with an efficient encryption scheme. DICOM images. Security and Communication Networks,
2017. https://doi.org/10.1155/2017/6729896
Chen, X., & Hu, C.-J. (2017). Adaptive medical image encryp­
tion algorithm based on multiple chaotic mapping. Saudi
Disclosure statement Journal of Biological Sciences, 24(8), 1821–1827. https://doi.
No potential conflict of interest was reported by the author(s). org/10.1016/j.sjbs.2017.11.023
Chen, Y., Tang, C., & Ye, R. (2020). Cryptanalysis and
improvement of medical image encryption using
high-speed scrambling and pixel adaptive diffusion. Signal
ORCID Processing, 167, 107286. https://doi.org/10.1016/j.sigpro.
Mohit Dua http://orcid.org/0000-0001-7071-8323 2019.107286
Choi, U. S., Cho, S. J., & Kang, S. W. (2019). Color medical
image encryption using 3D chaotic cat map and NCA. In
2019 10th IFIP International Conference on New
Data availability statement Technologies, Mobility and Security (NTMS). Canary
Data sharing does not apply to this article as no datasets were Islands, Spain, 1–5.
generated or analyzed during this study Dagadu, J. C., & Li, J. (2018). Context-based watermarking
cum chaotic encryption for medical images in telemedicine
applications. Multimedia Tools and Applications, 77(18),
References 24289–24312. https://doi.org/10.1007/s11042-018-5725-y
Dagadu, J. C., Li, J.-P., & Aboagye, E. O. (2019). Medical
Ahmad, J., & Hwang, S. O. (2016). A secure image encryption image encryption based on hybrid chaotic DNA diffusion.
scheme based on chaotic maps and affine transformation. Wireless Personal Communications, 108(1), 591–612.
Multimedia Tools and Applications, 75(21), 13951–13976. https://doi.org/10.1007/s11277-019-06420-z
https://doi.org/10.1007/s11042-015-2973-y Dagadu, J. C., Li, J., & Shah, F. (2017). An efficient di-chaotic
Akkasaligar, P. T., & Biradar, S. (2016). Secure medical image diffusion based medical image cryptosystem. In 2017 14th
encryption based on intensity level using chao’s theory and International Computer Conference on Wavelet Active
DNA cryptography. 2016 IEEE International Conference on Media Technology and Information Processing
Computational Intelligence and Computing Research (ICCWAMTIP). Chengdu, China, 206–210.
(ICCIC). Chennai, India, 1–6. Dagadu, J. C., Li, J.-P., Shah, F., Mustafa, N., & Kumar, K.
Akkasaligar, P. T., & Biradar, S. (2020). Selective medical (2016). DWT based encryption technique for medical
image encryption using DNA cryptography. Information images. In 2016 13th International Computer Conference
Security Journal, 29(2), 91–101. https://doi.org/10.1080/ on Wavelet Active Media Technology and Information
19393555.2020.1718248 Processing (ICCWAMTIP). Chengdu, China, 252–255.
Barik, R. C., & Changder, S. (2021). A novel and efficient Dai, Y., Wang, H., & Wang, Y. (2016). Chaotic medical image
amino acid codon based medical image encryption scheme encryption algorithm based on bit-plane decomposition.
colligating multiple chaotic maps. Multimedia Tools and International Journal of Pattern Recognition and Artificial
Applications, 80(7), 10723–10760. https://doi.org/10.1007/ Intelligence, 30(4), 1657001. https://doi.org/10.1142/
s11042-020-09930-2 S0218001416570019
Bigge, J., & Bohl, E. (1985). Deformations of the bifurcation Deb, S., & Bhuyan, B. (2021). Chaos-based medical image
diagram due to discretization. Mathematics of encryption scheme using special nonlinear filtering
356 S. PANKAJ AND M. DUA

function based LFSR. Multimedia Tools and Applications, Hajjaji, M. A., Dridi, M., & Mtibaa, A. (2019). A medical
80(13), 19803–19826. https://doi.org/10.1007/s11042-020- image crypto-compression algorithm based on neural net­
10308-7 work and PWLCM. Multimedia Tools and Applications, 78
Devi, A., Sharma, A., & Rangra, A. (2015). A review on DES, (11), 14379–14396. https://doi.org/10.1007/s11042-018-
AES and blowfish for image encryption & decryption. 6795-6
International Journal of Computer Science and Han, B., Jia, Y., Huang, G., & Cai, L. (2020). A medical image
Information Technologies, 6(3), 3034–3036. encryption algorithm based on hermite chaotic neural
Dou, Y., & Li, M. (2020). Cryptanalysis of a new color image network. In 2020 IEEE 4th Information Technology,
encryption using combination of the 1D chaotic map. Networking, Electronic and Automation Control
Applied Sciences (Switzerland), 10(6), 129–137. https://doi. Conference (ITNEC). Chongqing, China, 1, 2644–2648.
org/10.3390/app10062187 Hosseinzadeh, R., Zarebnia, M., & Parvaz, R. (2019). Hybrid
Elamir, M. M., Al-Atabany, W. I., & Mabrouk, M. S. (2021). image encryption algorithm based on 3D chaotic system
Hybrid image encryption scheme for secure E-health and choquet fuzzy integral. Optics and Laser Technology,
systems. Network Modeling Analysis in Health Informatics 120(June), 105698. https://doi.org/10.1016/j.optlastec.2019.
and Bioinformatics, 10(1), 1–8. https://doi.org/10.1007/ 105698
s13721-021-00306-6 Hua, Z., Yi, S., & Zhou, Y. (2018). Medical image encryption
El Assad, S., & Farajallah, M. (2016). A new chaos-based using high-speed scrambling and pixel adaptive diffusion.
image encryption system. Signal Processing: Image Signal Processing, 144, 134–144. https://doi.org/10.1016/j.
Communication, 41, 144–157. https://doi.org/10.1016/j. sigpro.2017.10.004
image.2015.10.004 Ibrahim, S., Alhumyani, H., Masud, M., Alshamrani, S. S.,
El-Deen, A., El-Badawy, E., & Gobran, S. (2014). Digital image Cheikhrouhou, O., Muhammad, G., Hossain, M. S., &
encryption based on RSA algorithm. Journal of electronics Abbas, A. M. (2020). Framework for efficient medical
and communication engineering, 9(1), 69–73. https://doi. image encryption using dynamic S-boxes and chaotic
org/10.9790/2834-09146973 maps. Institute of Electrical and Electronics Engineers
El-Shafai, W., Khallaf, F., El-Rabaie, E.-S. M., & Abd El- Access, 8, 160433–160449. https://doi.org/10.1109/
Samie, F. E. (2021). Robust medical image encryption ACCESS.2020.3020746
based on DNA-chaos cryptosystem for secure telemedicine Ismail, S. M., Said, L. A., Radwan, A. G., Madian, A. H., &
and healthcare applications. Journal of Ambient Intelligence Abu-Elyazeed, M. F. (2018). Generalized double-humped
and Humanized Computing, 12(10), 9007–9035. https://doi. logistic map-based medical image encryption. Journal of
org/10.1007/s12652-020-02597-5 Advanced Research, 10, 85–98. https://doi.org/10.1016/j.
Frank, G. W. (1990). Recovering the Lyapunov exponent from jare.2018.01.009
chaotic time series Accessed 12 10 2023 https://ir.lib.uwo. Jeevitha, S., & Amutha Prabha, N. (2020). Novel medical
ca/digitizedtheses/1998. image encryption using DWT block-based scrambling
Fu, C., Shan, Y.-F., He, M.-Y., Yu, Z.-Y., & Wu, H.-L. (2018). and edge maps. Journal of Ambient Intelligence and
A new medical image encryption algorithm using multiple Humanized Computing, 12(3), 3373–3388. https://doi.org/
1-D chaotic maps. In 2018 IEEE International Conference 10.1007/s12652-020-02399-9
on Systems, Man, and Cybernetics (SMC). Miyazaki, Japan, Kamal, S. T., Hosny, K. M., Elgindy, T. M., Darwish, M. M., &
2055–2060. Fouda, M. M. (2021). A new image encryption algorithm
Gafsi, M., Abbassi, N., Hajjaji, M. A., Malek, J., Mtibaa, A., & for Grey and color medical images. Institute of Electrical
Pan, Z. (2020). Improved chaos-based cryptosystem for and Electronics Engineers Access, 9, 37855–37865. https://
medical image encryption and decryption. Scientific doi.org/10.1109/ACCESS.2021.3063237
Programming, 2020, 1–22. https://doi.org/10.1155/2020/ Ke, G., Wang, H., Zhou, S., & Zhang, H. (2019).
6612390 Encryption of medical image with most significant bit
Guesmi, R., & Farah, M. A. B. (2021). A new efficient medical and high capacity in piecewise linear chaos graphics.
image cipher based on hybrid chaotic map and DNA code. Measurement, 135, 385–391. https://doi.org/10.1016/j.
Multimedia Tools and Applications, 80(2), 1925–1944. measurement.2018.11.074
https://doi.org/10.1007/s11042-020-09672-1 Kocarev, L., Jakimoski, G., Stojanovski, T., & Parlitz, U.
Gupta, K., Silakari, S., Gupta, R., & Khan, S. A. (2009). An (1998). From chaotic maps to encryption schemes. IEEE
ethical way of image encryption using ECC. In 2009 First International Symposium on Circuits and Systems (ISCAS)
International Conference on Computational Intelligence, Monterey, CA, USA, 4, 514–517.
Communication Systems and Networks. Indore, India, Kumar, M., Gupta, P., & Pandey, S. K. (2021). A new medical
342–345. image encryption algorithm based on the 1D logistic map
Hafsa, A., Gafsi, M., Malek, J., Machhout, M., & Vitiello, A. associated with pseudo-random numbers. Multimedia
(2021). FPGA implementation of improved security Tools and Applications, 80(1), 1–34. https://doi.org/10.
approach for medical image encryption and decryption. 1007/s11042-020-08904-8
Scientific Programming, 2021, 1–20. https://doi.org/10. Kumar, M., Iqbal, A., & Kumar, P. (2016). A new RGB image
1155/2021/6610655 encryption algorithm based on DNA encoding and elliptic
 INFORMATION SECURITY JOURNAL: A GLOBAL PERSPECTIVE 357

curve Diffie–Hellman cryptography. Signal Processing, 125, Parvees, M. Y. M., Samath, J. A., & Bose, B. P. (2016). Secured
187–202. https://doi.org/10.1016/j.sigpro.2016.01.017 medical images-a chaotic pixel scrambling approach.
Kumar, S., Panna, B., & Jha, R. K. (2019). Medical image Journal of Medical Systems, 40(11), 1–11. https://doi.org/
encryption using fractional discrete cosine transform with 10.1007/s10916-016-0611-5
chaotic function. Medical & Biological Engineering & Parvees, M. Y. M., Samath, J. A., & Bose, B. P. (2017). Medical
Computing, 57(11), 2517–2533. https://doi.org/10.1007/ images are safe–an enhanced chaotic scrambling approach.
s11517-019-02037-3 Journal of Medical Systems, 41(10), 1–16. https://doi.org/10.
Laiphrakpam, D. S., & Khumanthem, M. S. (2017). Medical 1007/s10916-017-0809-1
image encryption based on improved ElGamal encryption Parvees, M. Y. M., Samath, J. A., & Bose, B. P. (2018).
technique. Optik, 147, 88–102. https://doi.org/10.1016/j. Providing confidentiality for medical image—an enhanced
ijleo.2017.08.028 chaotic encryption approach. In Advances in big data and
Lakshmi, C., Thenmozhi, K., Rayappan, J. B. B., cloud computing (pp. 309–317). Springer.
Rajagopalan, S., Amirtharajan, R., & Chidambaram, N. Ponuma, R., & Amutha, R. (2018). Compressive sensing based
(2020). Neural-assisted image-dependent encryption image compression-encryption using novel 1D-Chaotic
scheme for medical image cloud storage. Neural map. Multimedia Tools and Applications, 77(15),
Computing and Applications, 33(12), 6671–6684. https:// 19209–19234. https://doi.org/10.1007/s11042-017-5378-2
doi.org/10.1007/s00521-020-05447-9 Prabhavathi, K., Sathisha, C. P., & Ravikumar, K. M. (2018).
Liu, H., Kadir, A., & Liu, J. (2019). Color pathological image Region of interest based selective medical image encryption
encryption algorithm using arithmetic over Galois field and using multi chaotic system. In International Conference on
coupled hyper chaotic system. Optics and Lasers in Electrical, Electronics, Communication Computer
Engineering, 122, 123–133. https://doi.org/10.1016/j.optla Technologies and Optimization Techniques, ICEECCOT
seng.2019.05.027 2017, 2018-Janua, 804–808. https://doi.org/10.1109/
Liu, J., Ma, Y., Li, S., Lian, J., & Zhang, X. (2018). A new simple ICEECCOT.2017.8284614
chaotic system and its application in medical image Raja, S. P. (2019). Joint medical image compression–encryp­
encryption. Multimedia Tools and Applications, 77(17), tion in the cloud using multiscale transform-based image
22787–22808. https://doi.org/10.1007/s11042-017-5534-8 compression encoding techniques. Sādhanā, 44(2), 28.
Marotto, F. R. (1979). Chaotic behavior in the Hénon map­ https://doi.org/10.1007/s12046-018-1013-9
ping. Communications in Mathematical Physics, 68(2), Ravichandran, D., Murthy, B. K., Fathima, V.,
187–194. https://doi.org/10.1007/BF01418128 Balasubramanian, S., Amirtharajan, R., &
Masood, F., Driss, M., Boulila, W., Ahmad, J., Rehman, S. U., Amirtharajan, R. (2021). An efficient medical image
Jan, S. U., Qayyum, A., & Buchanan, W. J. (2021). encryption using hybrid DNA computing and chaos in
A lightweight chaos-based medical image encryption transform domain. Medical & Biological Engineering &
scheme using random shuffling and XOR operations. Computing, 59(3), 589–605. https://doi.org/10.1007/
Wireless Personal Communications, 127(2), 1405–1432. s11517-021-02328-8
https://doi.org/10.1007/s11277-021-08584-z Ravichandran, D., Praveenkumar, P., Rayappan, J. B. B., &
Moafimadani, S. S., Chen, Y., & Tang, C. (2019). A new Amirtharajan, R. (2016). Chaos based crossover and muta­
algorithm for medical color images encryption using chao­ tion for securing DICOM image. Computers in Biology and
tic systems. Entropy, 21(6), 577. https://doi.org/10.3390/ Medicine, 72, 170–184. https://doi.org/10.1016/j.comp
e21060577 biomed.2016.03.020
Nematzadeh, H., Enayatifar, R., Motameni, H., Ravichandran, D., Praveenkumar, P., Rayappan, J. B. B., &
Guimarães, F. G., & Coelho, V. N. (2018). Medical image Amirtharajan, R. (2017). DNA Chaos blend to secure med­
encryption using a hybrid model of modified genetic algo­ ical privacy. IEEE Transactions on Nanobioscience, 16(8),
rithm and coupled map lattices. Optics and Lasers in 850–858. https://doi.org/10.1109/TNB.2017.2780881
Engineering, 110, 24–32. https://doi.org/10.1016/j.optla Reddy, V. P. K., & Fathima, A. A. (2016). Efficient encryption
seng.2018.05.009 technique for medical X-ray images using chaotic maps. In
OISHI, S., & Inoue, H. (1982). Pseudo-random number gen­ 2016 International Conference on Wireless
erators and chaos. IEICE TRANSACTIONS (1976-1990), 65 Communications, Signal Processing and Networking
(9), 534–541. (WiSPNET). Chennai, India, 783–787.
Parameshachari, B. D., Panduranga, H. T., & Samonas, S., & Coss, D. (2014). The CIA strikes back:
Naveenkumar, S. K. (2017). Partial encryption of medical Redefining confidentiality, integrity and availability in
images by dual DNA addition using DNA encoding. In security. Journal of Information System Security, 10(3),
2017 International Conference on Recent Innovations in 1551-0123. https://www.proso.com/dl/Samonas.pdf
Signal Processing and Embedded Systems (RISE). Bhopal, Sangavi, V., & Thangavel, P. (2020). An exotic
India, 310–314. multi-dimensional conceptualization for medical image
Pareek, N. K., & Patidar, V. (2016). Medical image protection encryption exerting Rossler system and Sine map. Journal
using genetic algorithm operations. Soft Computing, 20(2), of Information Security & Applications, 55, 55. https://doi.
763–772. https://doi.org/10.1007/s00500-014-1539-7 org/10.1016/j.jisa.2020.102626
358 S. PANKAJ AND M. DUA

Saraswathi, P. V., & Venkatesulu, M. (2017). A novel stream Thanki, R., & Kothari, A. (2021). Multi-level security of med­
cipher using pseudo random binary sequence generator for ical images based on encryption and watermarking for
medical image encryption. In 2017 International telemedicine applications. Multimedia Tools and
Conference on Trends in Electronics and Informatics Applications, 80(3), 4307–4325. https://doi.org/10.1007/
(ICEI). Tirunelveli, India, 425–429. s11042-020-09941-z
Selvi, C. T., Amudha, J., & Sudhakar, R. (2021). Medical image Wang, Q., Chen, X., Wei, M., & Miao, Z. (2016). Simultaneous
encryption and compression by adaptive sigma filterized encryption and compression of medical images based on
synorr certificateless signcryptive levenshtein optimized tensor compressed sensing with 3D Lorenz.
entropy-coding-based deep neural learning. Multimedia Biomedical Engineering Online, 15(1), 1–20. https://doi.
Systems, 27(6), 1059–1074. https://doi.org/10.1007/ org/10.1186/s12938-016-0239-1
s00530-021-00764-y Wolfgang, R. B., & Delp, E. J., III. (1998). Overview of image
Shafique, A., Ahmed, J., Rehman, M. U., & Hazzazi, M. M. security techniques with applications in multimedia
(2021). Noise-resistant image encryption scheme for medical systems. Multimedia Networks: Security, Displays,
images in the chaos and wavelet domain. Institute of Terminals, and Gateways. Dallas, TX, United States, 3228,
Electrical and Electronics Engineers Access, 9, 59108–59130. 297–308.
https://doi.org/10.1109/ACCESS.2021.3071535 Wu, Y., Noonan, J. P., & Agaian, S. (2011). NPCR and UACI
Shankar, K., Elhoseny, M., Chelvi, E. D., randomness tests for image encryption. Cyber Journals:
Lakshmanaprabu, S. K., & Wu, W. (2018). An efficient Multidisciplinary Journals in Science and Technology,
optimal key based chaos function for medical image Journal of Selected Areas in Telecommunications (JSAT), 1
security. Institute of Electrical and Electronics Engineers (2), 31–38.
Access, 6, 77145–77154. https://doi.org/10.1109/ACCESS. Yin, S., & Li, H. (2020). GSAPSO-MQC: Medical image
2018.2874026 encryption based on genetic simulated annealing parti­
Shen, Y., Tang, C., Xu, M., & Lei, Z. (2021). Optical selective cle swarm optimization and modified quantum chaos
encryption based on the FRFCM algorithm and face bio­ system. Evolutionary Intelligence, 14(4), 1817–1829.
metric for the medical image. Optics & Laser Technology, https://doi.org/10.1007/s12065-020-00440-6
138, 106911. https://doi.org/10.1016/j.optlastec.2020. Yoshida, T., Mori, H., & Shigematsu, H. (1983). Analytic
106911 study of chaos of the tent map: Band structures, power
Shokouh Saljoughi, A., & Mirvaziri, H. (2019). A new method spectra, and critical behaviors. Journal of Statistical
for image encryption by 3D chaotic map. Pattern Analysis Physics, 31(2), 279–308. https://doi.org/10.1007/
and Applications, 22(1), 243–257. https://doi.org/10.1007/ BF01011583
s10044-018-0765-5 Yun-Peng, Z., Wei, L., Shui-Ping, C., Zheng-Jun, Z., Xuan, N.,
Singh, V. K., Singh, P. K., & Rai, K. N. (2018). Image encryp­ & Wei-di, D. (2009). Digital image encryption algorithm
tion algorithm based on circular shift in pixel bit value by based on chaos and improved DES. In 2009 IEEE
group modulo operation for medical images. In 2018 4th International Conference on Systems, Man and
International Conference on Computing Communication Cybernetics. San Antonio, TX, USA, 474–479.
and Automation (ICCCA). Greater Noida, India, 1–7.
Zhang, J., Tan, J., & Cheng, Y. (2017). Medical image encryp­
Stalin, S., Maheshwary, P., Shukla, P. K., Maheshwari, M.,
tion algorithm based on chaotic function. In 2017 4th
Gour, B., & Khare, A. (2019). Fast and secure medical
International Conference on Information Science and
image encryption based on non linear 4D logistic map
Control Engineering (ICISCE). Changsha, China, 446–450.
and DNA sequences (NL4DLM_DNA). Journal of
Medical Systems, 43(8), 1–17. https://doi.org/10.1007/ Zhou, J., Li, J., & Di, X. (2020). A novel lossless medical
s10916-019-1389-z image encryption scheme based on game theory with
Tang, Z., Song, J., Zhang, X., & Sun, R. (2016). Multiple-image optimized ROI parameters and hidden ROI position.
encryption with bit-plane decomposition and chaotic Institute of Electrical and Electronics Engineers Access,
maps. Optics and Lasers in Engineering, 80, 1–11. https:// 8, 122210–122228. https://doi.org/10.1109/ACCESS.
doi.org/10.1016/j.optlaseng.2015.12.004 2020.3007550